arXiv:2609.28726
Associative Memory for Quantum Entangled States
Qiyuan Hu, Kartiek Agarwal, Ivar Martin
Abstract
We generalize the canonical Hopfield model of associative memory, which stores classical states of individual spins, to store entangled quantum states. The construction is based on truncated projector-sum Hamiltonian, and we explicitly consider a special case that stores dimer singlet coverings of a spin-1/2 system. We analyze the stability of the stored memories and find that the capacity -- the number of stable patterns -- scales faster with the system size than one would expect from a naive analogy with the standard Hopfield model, suggesting a possible quantum advantage.
AI Summary
Can an associative-memory Hamiltonian store entangled states rather than classical spin patterns? This paper constructs Hamiltonians by expanding a sum of memory-state projectors and retaining low-operator-weight terms. For product-state memories, this reproduces classical dense Hopfield models. The entangled example stores perfect matchings of $N$ spin-$1/2$ particles, with each matched pair forming a singlet. The main model contains products of two Heisenberg bond operators, hence four-spin interactions. Stored dimer states are energetically stable against all single-spin rotations; the relevant two-spin instabilities are singlet-bond rearrangements generated by exchange operators. A combinatorial signal-to-noise analysis gives a restoring contribution of order $N$ and crosstalk variance of order $K/N$ for $K$ memories. Concentration bounds and a union bound give a sufficient local-stability regime scaling as $K=O(N^3/\log N)$. Simulations show approximate collapse versus $K/N^3$, with a 5% instability threshold near $0.015N^3$ at accessible sizes. Higher-order, $2p$-spin models are estimated to support $O(N^{2p-1}/\log N)$ memories. The capacity exponent therefore matches that of classical dense memories at equal spin-interaction order. A perturbative hybridization diagnostic shows similar finite-size scaling. Small-system eigenstate calculations, restricted to total spin zero and using randomized memory weights, examine singlet overlaps and concurrence. These do not resolve the cubic capacity scaling. The logarithmic correction is also unresolved numerically. The study addresses static stability rather than demonstrated retrieval dynamics: dissipation, retrieval protocols, and storage of more general entangled states remain open.
Central claim may not hold: The paper establishes restricted local energetic stability and supporting static diagnostics, but not successful associative retrieval of an entangled memory from a corrupted input. The analytical capacity argument gives a sufficient low-load stability condition, not a matching upper bound establishing the maximum capacity as N^3/log N. The suggested capacity advantage over standard Hopfield memory compares four-spin quantum interactions with two-spin classical interactions, while the paper's classical four-spin benchmark already has the same capacity exponent. The stability proof drops the f−g contribution even though its secondary-memory fluctuations have the same asymptotic order as the retained a−b noise, leaving the stated probability bound incomplete.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 15 pages, 6 figures